No, but it is absorbing specific tasks, and asking the question by job title gives you a useless answer. Sorted by task, the pattern is consistent: anything with a defined output and fast feedback has been absorbed or will be, and anything that requires a claim about your business has not. The real change is that the bottleneck moves from producing ads to reviewing and approving them, which reshapes a creative team rather than shrinking it to nothing.
Task by task, what has actually been absorbed
Rows below are tasks, not roles, because no real job maps cleanly to one column. "Absorbed" means the machine version is good enough that doing it by hand is hard to justify.
| Task | Status in September 2026 | Who owns it now |
|---|---|---|
| Concept generation | Assisted. Volume is free, selection is not | Machine drafts, human selects |
| Scriptwriting, first draft | Absorbed | Machine |
| Script editing and claim wording | Human | Human |
| Storyboarding for internal sign-off | Absorbed | Machine |
| Casting and direction | Human | Human |
| Filming and capturing raw footage | Human, and rising in value | Human |
| Logging, transcribing and tagging footage | Absorbed completely | Machine |
| Shot selection for a variation | Absorbed, quality depends on tagging | Machine |
| Rough cut assembly | Absorbed | Machine |
| Fine cut, timing, the first two seconds | Contested. Machine gets close, human still wins on hooks | Human on the ads carrying real spend |
| Voiceover recording | Absorbed for performance work | Machine |
| Music selection from a licensed set | Absorbed | Machine |
| Captions and subtitling | Absorbed, needs proofreading on names and numbers | Machine, human proofread |
| Resizing to 9:16, 4:5, 1:1, 16:9 | Absorbed. Nobody should be doing this by hand | Machine |
| Localisation into other languages | Mostly absorbed. Native drafting works, idiom and regulated claims do not | Machine, native-speaker check |
| Colour and brand overlays | Absorbed from a locked template | Machine |
| Deciding the offer | Not absorbed, and not close | Human |
| Compliance and claim review | Human, every time | Human |
| Approval | Human, named individual | Human |
| Experiment design | Human. Tools report, they do not design | Human |
| Reading results and calling a winner | Assisted. The maths is automated, the call is not | Human |
| Deciding what to test next | Human | Human |
| Deciding what to film next quarter | Human, and this is now strategic work | Human |
Read the middle of that table rather than the ends. The absorbed tasks are the ones that used to fill a junior editor's day: logging, versioning, resizing, subtitling, rough assembly. The unabsorbed ones are the ones that used to feel like overhead: writing constraints down, designing a test, deciding what to shoot. Overhead is now the job.
Where the bottleneck moves
Production used to be the constraint because every variation cost somebody a morning. Now a batch of 20 exists in minutes, and nothing about your review capacity changed.
Do the arithmetic. A careful pass over 20 variations takes a person 30 to 45 minutes with a checklist: claims, prices, dates, product accuracy, on-screen text, brand basics, and whether the six ideas are actually six ideas. Ship 200 variations a month and you have created five to eight hours a week of review that did not exist before, and it cannot be sampled, because each output was decided separately rather than stamped from one template. Approving variation three tells you very little about variation seventeen.
That is the whole shape of the change. Volume requirements are real: our 2026 fatigue benchmark puts week three at 45 to 70 percent below the launch baseline, most creative effectively dead within three weeks, a typical active campaign needing 8 to 20 live variations, and brands shipping 15 to 50 variants a month seeing 3 to 5 times longer campaign lifespan than quarterly refreshers. Its own conclusion is that throughput rather than talent is the limit. Tooling removed the throughput limit and handed you a review queue instead.
Two failure modes follow, and we see both. Teams that keep the old review process drown, then quietly cut output back to what they can inspect, which returns them to where they started. Teams that abandon review ship a wrong price or an uncleared claim, get a campaign pulled, and learn the same lesson expensively. The version that works is a fixed checklist, batch-level review, and constraints moved out of review notes and into the brief, which is the single change that most reduces review load. Our agency creative workflow writeup has the process we would run across multiple clients with different rule sets.
The skills that gain value
Five, and none of them are software skills.
Writing taste down. The scarce ability is turning "this feels off-brand" into a constraint a machine can obey: no ironic register, never the word premium, no price on screen. People who can articulate a house style as rules become the highest-leverage members of a creative team, because their judgement now applies to every future batch instead of one edit.
Experiment design. Most creative tests are underpowered and get called anyway. Knowing what volume you need before believing a 12 percent lift, and being willing to report that a test proved nothing, is worth more than it used to be, because you now have more tests to be wrong about.
Deciding what to film. Machine assembly is bounded by your library. Whoever can look at a quarter of results and say "we need twenty seconds of unglamorous product-in-use coverage and two more environments" improves every variation produced afterwards. This is the most undervalued skill in the new setup.
Compliance literacy. Knowing which claims need substantiation, which markets need which disclaimer, and which categories cannot use before-and-after imagery. A generator will not warn you.
Hook craft. The first two seconds still reward human judgement more than any other part of the ad, and typical hook rates of 20 to 30 percent on TikTok in-feed and 15 to 25 percent on Meta feed leave plenty of room to earn your salary there.
The skills that lose value are the mechanical ones: manual resizing, manual subtitling, manual versioning, building the fifteenth cut of an approved edit. If that was most of a role, the role changes.
What this looks like in headcount terms
We are not going to pretend nothing happens. Here is our honest read.
A three-person in-house team that was one strategist and two editors tends to become one strategist, one editor who spends most of the week reviewing and fixing hooks rather than assembling, and a real budget line for shoot days. Same headcount, different allocation, and more raw footage entering the library, which is the input that compounds.
The roles genuinely exposed are junior production roles whose day is versioning and resizing, and freelance work priced per deliverable. If your invoice says "10 cutdowns", that line item is under pressure. Roles that grow are creative operations, which barely existed as a title three years ago and is now the function that owns the review queue, plus anyone who shoots.
For agencies the change is a pricing problem before it is a staffing one. A retainer justified by output volume gets harder to defend when output volume is cheap, so the defensible retainer is priced on judgement: test design, claim clearance, what to shoot, what to kill. Some agencies will not make that transition, and it will look like AI replacing them when it was really a business model that stopped matching the cost of the work.
What we have not seen anywhere is a team that removed the human and kept the results. The definition we work from is in our glossary entry on what an AI agent for video ads is, and the practical integration notes are in our piece on running an ad agent inside an existing stack.
Disclosure: Genyad is our product, so we have an obvious interest here. It assembles variations from footage you already own. It has no AI avatars or synthetic presenters, no static banners, no product-URL import, no product-feed rendering, no predicted performance scores and no direct publishing to Meta or TikTok. It does not shoot footage, decide your offer, or approve anything.
Frequently asked questions
Which creative roles are most exposed to AI?
Roles whose week is mostly mechanical production: logging, versioning, resizing, subtitling and assembling cutdowns of an approved edit. Roles centred on judgement, claim clearance, experiment design and deciding what to shoot are gaining value, and the same person often holds both kinds of task. The exposure is to the task mix, not the job title.
Do teams using AI creative tools actually get smaller?
In our experience they stay the same size and spend their time differently, with production hours falling and review hours rising. Teams that cut headcount to match the production saving usually find they cannot approve what they can now produce, so output quietly falls back to the old level. Budget freed from production tends to go into filming more raw material.
If AI can write scripts, do I still need a copywriter?
Yes, for the parts that decide whether the ad is true and whether it argues the right thing. First drafts and volume variants are genuinely absorbed, so the work shifts to writing the brief, wording claims defensibly, and editing hooks. That is a smaller amount of typing and a larger amount of responsibility.
What is the new bottleneck if it is not production?
Review and approval. A careful pass over 20 variations takes 30 to 45 minutes and cannot be sampled the way template renders can, because each output involved separate decisions. Teams that do not redesign review around a checklist end up capping their own output at whatever they can inspect.